Ready-to-eat food environments and risk of incident heart failure: a prospective cohort study

Abstract

Background:

Food environments have been linked to cardiovascular diseases; however, few studies have assessed the relation between food environments and the risk of heart failure (HF). We aimed to evaluate the association between ready-to-eat food environments and incident HF at individual-level in a large prospective cohort.

Methods:

Exposure to ready-to-eat food environments, comprising pubs or bars, restaurants or cafeterias, and fast-food outlets, were individually measured as both proximity and density metrics. We also developed a composite ready-to-eat food environments density score by summing the densities of three types of food environments. Cox proportional analyses were applied to assess the associations of each single type and the composite food environments with HF risk.

Results:

Closer proximity to and greater density of ready-to-eat food environments, particularly for pubs and bars and fast-food outlets (P<0.05 for both proximity and density metric) were associated with an elevated risk of incident HF. Compared to those with no exposure to composite ready-to-eat food environments, participants in the highest density score category had a 16% (8%−25%, P<0.0001) higher risk of HF. In addition, we found significant interactions of food environments with education, urbanicity, and density of physical activity facilities on HF risk (all P_interaction <0.05), the ready-to-eat food environments-associated risk of HF was stronger among participants who were poorly educated, living in urban areas, and without physical activity facilities.

Conclusions:

Exposure to ready-to-eat food environments is associated with a higher risk of incident HF, suggesting the potential importance of minimizing unfavorable food environments in the prevention of HF.

Introduction:

Heart failure (HF) remains a major clinical and public health problem^1–4 and the prevalence of HF exhibits considerable geographic disparities across various populations.^1,2,5,6 Rapid modernization and urbanization lead to great shifts in diets, referred to as nutrition transition, and have been recognized as key contributors to non-communicable diseases including HF.^7,8 Underlying this transition is the change in the food environments that promotes a convenient ready-to-eat meal lifestyle against traditional home-prepared food.^9–11 Growing evidence has also shown that neighborhood food environment may shape an individual’s eating behaviors and dietary patterns,^12,13 which have emerged as important drivers in the development of HF.^14–18

Several previous studies have linked food environment exposure with HF as a component of composite cardiovascular events;^19,20 however, studies on the direct relationship between exposure to food environments and HF risk are limited. Only one prior study has investigated the link of incident HF with residential fast-food environments, where elevated fast-food density was associated with an increased HF risk.²⁰ Notably, the previous study had a short follow-up of one year. Furthermore, the previous study only analyzed the density measure of the specific fast-food environment, whereas the proximity measures and other types of ready-to-eat food environments (e.g., pubs or bars and restaurants or cafeterias) were not assessed. In addition, no study has assessed the relation between the overall density of ready-to-eat food environments combining various types of food environments with HF risk.

In this study, we prospectively analyzed the relationship between exposure to ready-to-eat food environments and incident HF risk by using a comprehensive evaluation of accessibility to food environments, including both proximity and density metrics for various ready-to-eat food environments, in the large UK Biobank cohort. We also tested the modification effects sex, urbanicity, education, and density of formal physical activity facilities.

Method:

The authors declare that all supporting data are available within the article and its online supplementary files.

Study design and participants

The UK Biobank is a population-based prospective cohort that includes more than 0.5 million community-dwelling individuals, aged 37–73 years, recruited from 22 assessment centers across England, Scotland, and Wales between March 13, 2006, and October 1, 2010. The study design and methods were described in detail elsewhere.²¹ Written informed consent was received from all participants before enrolment in the study. The UK Biobank study was approved by both the National Health Service National Research Ethics Service (Ref: 11/NW/0382) and the Institutional Review Board of Tulane University (2018–1872).

We used participants from the baseline assessment of the UK Biobank cohort. Among the 502 505 participants with available data in the current study, we excluded those with HF (n= 2,419) at baseline. This study was restricted to 478,598 participants who had complete data for exposure to food environments.

Assessment of ready-to-eat food environments

Individual-level exposure to ready-to-eat food environments was derived from the linked UK Biobank Urban Morphometric Platform (UKBUMP). UKBUMP is a high-resolution spatial database of objectively measured built environment metrics surrounding each participant’s residential address, deriving from multiple national-level spatial datasets.²² These metrics included proximity to and density of various health-related land uses and services (e.g., food environments), street network accessibility, physical environment (such as buildings, greenness, and terrain), and pollution.

Accessibility to ready-to-eat food environments was measured as both proximity (distance, m) to and density (count, units/km²) of pubs or bars (OS land use code CR06), restaurants or cafeterias (CR07), and fast-food outlets or hot and cold takeaways (CR10) within the defined 1-km street-network buffer of each participant’s place of residence. In concordance with previous analyses,^23,24 the 1-km street buffer areas were used as functional street neighborhoods corresponding to about 10–15 min of walking at a moderate pace. Destination proximity measures defined as street network distances from the residential address to the nearest ready-to-eat food destinations were measured by contrast with the density measure to check whether our results were sensitive to the measure of the food environment. Although proximity and density measures of the food environment are correlated, they are nonetheless theoretically distinct constructs.^25,26

In addition to each single-type of ready-to-eat food environment measure, we generate a composite ready-to-eat food environments density score from the summed densities of all three types of ready-to-eat food outlets (pubs and bars, restaurants and cafes, and fast-food outlets or hot and cold takeaways) within the 1-km buffer area.

Moreover, we considered the potential confounding effects of other local environments. The neighborhood residential density (OS land use code RD*, log-transformed count of residential dwellings within a 1-km street- network buffer of home address), as well as the densities of parks (LP01–04) and physical activity facilities (CL06), were employed as a proxy for other neighborhood resources that will be correlated with the ready-to-eat food environments. The densities of petrol filling stations (OS land use code CR05) and train and bus stations or terminals (CT08) were also used as a proxy for potential exposures to other food-related retail outlets clustering around them.

Assessment of covariates:

In UK Biobank, participants completed extensive touchscreen questionnaires on sociodemographic characteristics, socioeconomic variables, lifestyle habits, medical history, and treatments at baseline. Various physical measurements were also collected and assessed according to well-defined and standardized protocols.²⁷

Sociodemographic and socioeconomic characteristics at recruitment included age, sex, race, Townsend deprivation index, urbanicity, education, employment status, and average total annual household income. Smoking status was defined based on self-reported information as ‘never’, ‘previous’, or ‘current ‘smokers. Alcohol intake was evaluated by the touchscreen questionnaire and reported as ‘never’, ‘special occasions only’, ‘1–3 times per month’, ‘once or twice a week’, ‘3 or 4 times a week’, and ‘daily or almost daily’. A healthy diet score was adapted from the AHA Guidelines and defined as adherence to five components of the followings: 1) Total fruit intake ≥4.5 pieces/week; 2) Total vegetable intake ≥4.5 servings/week (3 tablespoons of vegetable considered as 1 serving); 3) Total fish intake ≥2 servings/week; 4) Processed meat intake less than twice/week; 5) Red meat intake ≤5 times/week.²⁸ Physical activity was measured at baseline based on self-reported short-form International Physical Activity Questionnaire guidelines. The final physical activity volume was computed in total metabolic equivalent task minutes per week for three types of activity including walking, moderate, and vigorous activity. Weight and height were assessed at baseline and body mass index was calculated with weight (kg)/height (m²). Systolic and diastolic blood pressure were calculated from two automated or two manual measures of BP. Medications to treat high blood pressure, high cholesterol, and diabetes were assessed through touch-screen questionnaires.

Assessment of outcomes

All participants were followed up for health outcomes through linkage to national electronic health-related data sets, including Hospital Episode Statistics for England, Scottish Morbidity Record for Scotland, and Patient Episode Database for Wales. The primary outcome of interest was the incident risk of HF. Baseline prevalent HF was defined based on self-reported information and hospital inpatient records by ICD 9 code 428.0, 428.1, and ICD 10 code I50.0, I50.1, or I50.9 before the baseline recruitment date. Incidence of heart failure was collected using the follow-up data till May 2021 and defined based on International Classification Diseases edition 10th (ICD 10: I11.0, I13.0, I13.2, I50.X). Follow-up time was calculated for each participant as the time starting from the recruitment date to the date of the first diagnosis of HF or the censoring date (loss to follow-up or death), whichever came first. Censoring dates for mortality outcomes and hospitalization records were May 31, 2021.

Statistical analysis

Based on the existing skewed distribution of exposures, each density of the single-type food environment variable was first divided into 3 categories (0, 1–2, and 3 or more). Each destination distance of the food environment was categorized as living closer than 500m, 500–999m, 1000–1999m, or at least 2000 m from the nearest food environment. Composite ready-to-eat food environments density score was first divided into quintiles and then rearranged into a new five-category, where the first category represented no exposure inside the first quintile, the second category was created by combining the first quintile’s non-zero exposures with the second quintile’s original exposures, and the third to fifth categories represented the same exposures as the third to fifth quintiles.

Baseline characteristics of participants were summarized as mean (SD), median (Q₁-Q₃), and percentages for continuous and categorical variables. The missing indicator category for categorical variables and imputed mean values for continuous variables were used if there were missing values for covariates in the analysis.

Cox proportional analyses were performed to examine the associations between ready-to-eat food environments and incident risk of HF. Initially, each single type of ready-to-eat food environments in proximity and density was evaluated separately with incident HF. We sequentially adjusted for potential confounding covariates based on a priori literature, including sociodemographic (age, sex, race; model 1), neighborhood and individual socioeconomic variables (urbanicity, Townsend deprivation index, income, education, and employment status; model 2), lifestyle variables and medication treatments (smoking status, alcohol intake frequency, physical activity, healthy diet score, body mass index, blood pressure treatment, insulin treatment, lipid treatment), and other environment-related variables (nearby location of petrol-filling stations or train or bus stations or terminals, residential density, density of parks and physical activity facilities) (model 3). We then re-performed the above models with the composite ready-to-eat food environments density score variable. We also examined whether the associations between proximity to and density of ready-to-eat food environments and HF varied by sex, urbanicity, education, and density of physical activity facilities by introducing interaction terms in the fully adjusted models. We further evaluated for potential non-linearity in the associations between each single type and the composite food environment measures and risk of HF and plotted the dose-response curves using restricted cubic spline models with Harrel’s knots. The proportional hazards assumption was evaluated using Schoenfeld residuals(P>.05 for all Schonfeld global tests).

We conducted several sensitivity analyses to evaluate the robustness of our results. First, we additionally adjusted for history of hypertension, hyperlipidemia, and type 2 diabetes at baseline. Second, we restricted the analysis to participants free of cardiovascular disease at baseline. Furthermore, we assessed the associations between ready-to-eat food environments and incident HF among participants who had lived in the current address for more than 5 years (around 85% stability). We repeated the analyses in the fully adjusted model.

All statistical analyses were performed with SAS 9.4 (SAS Institute Inc). All tests were two-sided and p < 0.05 was considered statistically significant.

Result

The baseline demographic and built environmental characteristics of the 478,598-study population are presented in Table 1. In total, the mean age of the participants included in the analysis was 56.54 (SD=8.09) years, 261,351 (54.31%) were female, and 451,046 (94.24%) were white European. Within the 1-km street buffer areas of the participants’ living locations, the median density of pubs and bars, restaurants and cafeterias, fast-food outlets, and composite ready-to-eat food environment were 1.29 units/km² [0–3.05], 1 unit/km² [0–4], 0 units/km² [0–1.81], and 3.57 units/km² [1–9.06], respectively; whereas the median street distances to these food environments were 692.23 m [401.33–1160.66] to pubs and bars, 820.5 m [465.46–1385.35] to restaurants and cafeterias, and 1135.93 m [615.59–2197.71] to fast-food outlets. 95,726 (20.00%) participants included in the analysis were exposed to the highest density category of composite ready-to-eat food environments (Table S1).

Table 1.

Baseline characteristics of UK Biobank participants.

	Baseline
No of participants	478598
Age, years	56.5[8.1]
Women	261,351(54.61%)
Race:
White European	451,046(94.24%)
Mixed	2,715(0.57%)
Asian	10,845(2.27%)
Black	7,319(1.53%)
Others	4,144(0.87%)
Missing	2,529(0.53%)
Household income (£):
<18 000	92,561(19.34%)
18 000–30 999	104,140(21.76%)
31 000–51 999	106,877(22.33%)
52 000–100 000	83,088(17.36%)
>100 000	21,672(4.53%)
missing	70,260(14.68%)
Employment status
Paid employment or self-employed	273,924(57.23%)
Retired	159,592(33.35%)
Unable to work	15,473(3.23%)
Unemployed	7,699(1.61%)
Home duties, carer, student, volunteer, or other	16,616(3.47%)
missing	5,294(1.11%)
College or University degree, yes	153,155(32.00%)
Townsend deprivation index	−1.38[3.03]
Urbanicity
Urban	408,739(85.40%)
Non-urban	65,598(13.71%)
Data missing	4,261(0.89%)
Smoking status:
Never	261,692(54.68%)
Former	164,447(34.36%)
Current	49,750(10.39%)
Missing	2,709(0.57%)
Alcohol intake:
Daily or almost daily	96,582(20.18%)
Three or four times a week	110,307(23.05%)
Once or twice a week	123,839(25.88%)
One to three times a month	53,202(11.12%)
Special occasions only	55,075(11.51%)
Never	38,252(7.99%)
Missing	1,341(0.28%)
Healthy diet score	2.39[0.89]
Physical activity, METs	2655.51[2439.05]
BMI, kg/m2	27.43[4.78]
SBP, mmHg	137.98[18.62]
DBP, mmHg	82.31[10.11]
Drug use:
Lipid treatment	82,211(17.18%)
Insulin treatment	5,176(1.08%)
Blood pressure treatment	98,863(20.66%)
Distance to nearest food environment, m *
Pubs and bars	692.23[401.33–1160.66]
Restaurants and cafeterias	820.5[465.46–1385.35]
Fast food outlets or hot and cold takeaways	1135.93[615.59–2197.71]
Density of food environment (units/km2) †
Pubs and bars	1.29[0–3.05]
Restaurants and cafeterias	1[0–4]
Fast food outlets or hot and cold takeaways	0[0–1.81]
Composite ready-to-eat food environments density score ‡	3.57[1–9.06]
Density of physical activity environment †	1.29[0–3]
0	149,344(31.20%)
1	131,985(27.58%)
2 or more	197,269(41.22%)
Range	0 to 38.88
Density of parks and other public/open green spaces †	0.81[0–2.16]
0	216,923(45.32%)
1 or 2	165,379(34.55%)
3 or more	96,296(20.12%)
Range	0 to 83.16
Residential density †	1986.21[1285.29]
Density of petrol-filling stations †	0.4[0.67]
Density of train and bus stations or terminals †	0.18[0.54]

Data are mean [SD], median [Q₁-Q₃], or N (%). MET: metabolic equivalent of task; BMI, Body mass index; DBP, diastolic blood pressure; SBP, systolic blood pressure.

^*Distance to nearest facilities in 1000 m buffer.

^†Number of facilities in 1000 m buffer.

^‡The density of composite ready-to-eat food environments is the density of pubs and bars, restaurants and cafeterias, and hot and cold takeaways, presented as units/km² within a 1-km street catchment area.

We documented a total of 12,956 incident cases of HF during a median of 12.23 years of follow-up. Table 2 illustrates the risk of HF in associations with proximity to ready-to-eat food environments. The decrease in street distance to the nearest ready-to-eat food environments was associated with the increased risk of HF. Among the three types of ready-to-eat food environments, the results remained significant for proximity to pubs and bars, and proximity to fast-food outlets, but not proximity to restaurants and cafeterias, in models 2–3 with additional adjustment for confounding effect. Participants resident in places within the closest distance category of ready-to-eat food environments (<500 m) had a higher risk of HF (for pubs and bars, adjusted HR (AHR)=1.13 [1.05, 1.22], P=0.0007; and for fast-food outlets, AHR=1.10 [1.04, 1.16], P=0.0017) than those in the reference category with the farthest distance (>=2000m). However, no significant differences were observed between the reference (farthest) and the second farthest distance (1000m-2000m) categories on the risk of HF.

Table 2.

Associations between the proximity (distance) to nearest ready-to-eat food environments and the risk of heart failure in UK Biobank participants.

		n cases/n total	Model 1			Model 2			Model3
			AHR (95% CI)	P1	P trend1	AHR (95% CI)	P2	P trend2	AHR (95% CI)	P3	P trend3
Pubs and bars	<500 m	4667/162903	1.30 [1.21, 1.39]	<.0001	<.0001	1.11 [1.04, 1.19]	0.0026	0.0003	1.13 [1.05, 1.22]	0.0007	<.0001
	500 m to 999 m	4547/164825	1.20 [1.12, 1.28]	<.0001		1.11 [1.04, 1.19]	0.0024		1.13 [1.05, 1.21]	0.0008
	1000 m to 1999 m	2700/107748	1.08 [1.00, 1.16]	0.047		1.04 [0.97, 1.12]	0.2702		1.05 [0.97, 1.13]	0.2061
	≥2000 m	1042/43122	1(ref)			1(ref)			1(ref)
Restaurants and cafeterias	<500 m	3658/132449	1.22 [1.15, 1.30]	<.0001	<.0001	0.99 [0.93, 1.06]	0.7764	0.8999	0.98 [0.92, 1.06]	0.6233	0.7838
	500 m to 999 m	4272/154074	1.16 [1.10, 1.23]	<.0001		1.01 [0.95, 1.07]	0.7855		1.00 [0.93, 1.06]	0.8939
	1000 m to 1999 m	3524/131137	1.10 [1.04, 1.17]	0.0019		0.99 [0.93, 1.06]	0.837		0.98 [0.92, 1.05]	0.6097
	≥2000 m	1502/60938	1(ref)			1(ref)			1(ref)
Fast food outlets or hot and cold takeaways	<500 m	2624/88285	1.36 [1.29, 1.43]	<.0001	<.0001	1.09 [1.03, 1.15]	0.0016	<.0001	1.10 [1.04, 1.16]	0.0017	0.0002
	500 m to 999 m	3617/124071	1.27 [1.21, 1.33]	<.0001		1.11 [1.06, 1.17]	<.0001		1.12 [1.06, 1.18]	<.0001
	1000 m to 1999 m	3512/132733	1.14 [1.09, 1.20]	<.0001		1.04 [0.99, 1.10]	0.1092		1.05 [1.00, 1.11]	0.0525
	≥2000 m	3203/133509	1(ref)			1(ref)			1(ref)

AHR, adjusted hazard ratio; CI, confidence interval; Distance to nearest food environments is presented as meters within a 1-km street buffer size.

Model 1: adjusted for age, sex, race.

Model 2: model 1 + adjusted for urban or non-urban status, Townsend deprivation index, individual socioeconomic characteristics (income, education, and employment status).

Model 3: model 2 + adjusted for alcohol drinking frequency, smoking status, physical activity, healthy diet score, body mass index, blood pressure treatment, insulin treatment, lipid treatment, residential density, presence of bus or train stations, presence of petrol-filling stations, density of parks and physical activity facilities.

The results examining the associations between the density of different types of ready-to-eat food environments and HF are shown in Table 3. We found elevated risks of HF across the higher density categories of pubs and bars and fast-food outlets (all P<0.05). Additional adjustment for covariates attenuated these associations, which remained significant across models 1–3. Compared with participants with no exposure to pubs and bars and fast-food outlets, those in the highest density category (>=3 units/km²) showed a significantly higher risk of HF (AHR= 1.14 [1.09, 1.20], P<.0001 for pubs and bars; AHR=1.12 [1.07, 1.18], P<.00001 for fast-food outlets) in the fully adjusted model 3. Exposure to restaurants and cafeterias was associated with a higher risk of HF in model 1, however, no significant association was observed after further adjustment for other confounding variables in models 2 and 3.

Table 3.

Associations of the density of different types of ready-to-eat food environments and the composite ready-to-eat food environments density score with the risk of heart failure in UK Biobank participants.

		n cases/n total	Model 1			Model 2			Model3
			AHR (95% CI)	P1	P trend1	AHR (95% CI)	P2	P trend2	AHR (95% CI)	P3	P trend3
Pubs and bars	0	3616/146048	1(ref)		<.0001	1(ref)		<.0001	1(ref)		<.0001
	1 or 2	5414/200458	1.12 [1.07, 1.16]	<.0001		1.06 [1.02, 1.11]	0.0054		1.07 [1.02, 1.12]	0.0028
	3 or more	3926/132092	1.29 [1.24, 1.35]	<.0001		1.11 [1.06, 1.16]	<.0001		1.14 [1.09, 1.20]	<.0001
Restaurants and cafeterias	0	4892/187283	1(ref)		<.0001	1(ref)		0.6052	1(ref)		0.6586
	1 or 2	4065/142864	1.12 [1.07, 1.16]	<.0001		1.02 [0.98, 1.07]	0.2989		1.01 [0.97, 1.06]	0.6582
	3 or more	3999/148451	1.11 [1.07, 1.16]	<.0001		0.99 [0.94, 1.03]	0.5519		0.99 [0.94, 1.04]	0.6271
Fast food outlets or hot and cold takeaways	0	6572/261525	1(ref)		<.0001	1(ref)		<.0001	1(ref)		<.0001
	1 or 2	4076/141190	1.19 [1.14, 1.24]	<.0001		1.06 [1.02, 1.11]	0.0022		1.06 [1.02, 1.11]	0.0054
	3 or more	2308/75883	1.29 [1.23, 1.36]	<.0001		1.11 [1.06, 1.16]	<.0001		1.12 [1.07, 1.18]	<.0001
Composite ready-to-eat food environments density score	Quintile 1 (0.0–0.0)	1884/80659	1(ref)		<.0001	1(ref)		0.0002	1(ref)		<.0001
	Quintile 2 (0.4–2.3)	2860/110763	1.10 [1.04, 1.17]	0.0011		1.03 [0.97, 1.09]	0.3052		1.03 [0.97, 1.09]	0.3238
	Quintile 3 (2.3–5.0)	2708/95724	1.25 [1.17, 1.32]	<.0001		1.10 [1.04, 1.17]	0.0017		1.11 [1.04, 1.18]	0.0011
	Quintile 4 (5.0–11.2)	2760/95726	1.29 [1.21, 1.37]	<.0001		1.10 [1.03, 1.16]	0.0028		1.11 [1.04, 1.18]	0.002
	Quintile 5 (11.2–383.0)	2744/95726	1.37 [1.29, 1.45]	<.0001		1.11 [1.04, 1.18]	0.0016		1.16 [1.08, 1.25]	<.0001

AHR, adjusted hazard ratio; CI, confidence interval; Density of food environments is presented as units/km² within a 1-km street buffer size; The composite ready-to-eat food environments density score is the density of pubs and bars, restaurants and cafeterias, and fast-food outlets or hot and cold takeaways, presented as units/km².

Model 1: adjusted for age, sex, race.

Model 2: model 1 + adjusted for urban or non-urban status, Townsend deprivation index, individual socioeconomic characteristics (income, education, and employment status).

Model 3: model 2 + adjusted for alcohol drinking frequency, smoking status, physical activity, healthy diet score, body mass index, blood pressure treatment, insulin treatment, lipid treatment, residential density, presence of bus or train stations, presence of petrol-filling stations, density of parks and physical activity facilities.

In analyses of the composite ready-to-eat food environments density score, participants in the highest density category (>11.2 units/km²) showed a higher risk of HF than those in the reference category who had no exposure to ready-to-eat food environments (AHR=1.16 [1.08, 1.25], P<0.0001 in model 3; Table 3).

Sensitivity analyses to further adjust for history of hypertension, hyperlipidemia, and type 2 diabetes (Table S2), and to re-perform the fully adjusted models among a subset of participants free of cardiovascular diseases (Table S3) or living in the current address for at least five years(Table S4) showed robust and consistent results, indicating a higher risk of HF among participants resident in places with closer proximity to and greater density of pubs and bars and fast-food outlets, and greater density of composite ready-to-eat food environments.

We further evaluated the non-linear association of continuous measures of food environments and HF. We found significant non-linear associations of the density of composite ready-to-eat food environments (P=0.002) and pubs and bars (P=0.0002) on HF with Harrel’s knots in our fitted restricted cubic spline models (Figure 1 and Figure S1–B). The risks of HF were attenuated beyond the threshold of density 13 units/km² for composite ready-to-eat food environments and 4 units/km² for pubs and bars. Test for the nonlinearity of the relationship between closer proximity to pubs and bars and HF (P=0·321) was however not significant with Harrel’s knots (Figure S1–A). Specifically, the risk of HF increased along with the decrease in distance to pubs and bars. A consistent linear association between either proximity to or density of fast-food outlets and HF was observed (Figure S1–E&F). Closer distance to and greater density of fast-food outlets were associated with substantially increased risk of HF, with no evidence against non-linearity. There was no sufficient evidence of associations between the proximity to and density of restaurants and cafeterias and risk of HF.

Figure 1. Association of the composite ready-to-eat food environments density score with risk of heart failure, allowing for non-linear effects.

The continuous line represents the estimated adjusted hazard ratio of heart failure and dashed lines represent 95% CIs. Model was fitted for HF, with restricted cubic splines with Harrell’s knots, adjusting for age, sex, race, urban or non-urban status, Townsend deprivation index, individual socioeconomic characteristics (income, education, and employment status), alcohol drinking frequency, smoking status, physical activity, healthy diet score, body mass index, blood pressure treatment, insulin treatment, lipid treatment, residential density, presence of bus or train stations, presence of petrol-filling stations, density of parks and physical activity facilities. The reference level was zero for the density of composite ready-to-eat environments with adjusted hazard ration fixed as 1.

In the subgroup analyses, food environments-associated risk of HF did not differ by sex. We found statistical evidence of modification effects of education (P_interaction =0.0183) and density of physical activity facilities (P_interaction =0.0038) with the association between composite ready-to-eat food environments density score and HF (Figure 2). Effect sizes were comparatively greater in the higher density score category of composite ready-to-eat food environments among those without college or University degree, and those with no exposure to formal physical activity facilities. For the association between the proximity to and density of fast-food outlets and HF, we found evidence of effect modification by urbanicity (with proximity P_interaction =0.0345; with density P_interaction =0.0182), with positive and stronger associations among participants living in urban areas (Figure S2).

Figure 2. Association between the composite ready-to-eat food environments density score and risk of heart failure, stratified by education and density of physical activity facilities.

Results were fully adjusted for age, sex, race, urban or non-urban status, Townsend deprivation index, individual socioeconomic characteristics (income, education, and employment status), alcohol drinking frequency, smoking status, physical activity, healthy diet score, body mass index, blood pressure treatment, insulin treatment, lipid treatment, residential density, presence of bus or train stations, presence of petrol-filling stations, density of parks and physical activity facilities. AHR, adjusted hazard ratio; CI, confidence interval.

Discussion

In the large prospective cohort UK Biobank, we found consistently positive associations of both proximity to and density of ready-to-eat food environments with the risk of incident HF. Participants resident in places with closer proximity to and greater density of ready-to-eat food environments had a 10% to 16% elevated risk of HF than those without, particularly for the types of pubs and bars and fast-food outlets. These results remained unchanged even after adjusting for individual-level demographic, lifestyle, and environmental confounders. In addition, we found the association of ready-to-eat food environments with incident HF was modified by education, urbanicity and density of physical activity facilities.

To the best of our knowledge, this is the first prospective study to assess the relation between ready-to-eat food environments and incident HF risk with long-term follow-up. Our findings were partly supported by previous studies which showed the positive associations between food environments and cardiovascular outcomes in both individual and neighborhood area-level. Among a nationwide sample in the Netherlands with one-year follow-up, a significantly increased incidence of HF was found among adults with elevated fast-food outlets within 1 km of the residential address, though the evidence was more pronounced for urban areas and for cardiovascular disease or coronary heart disease.²⁰ Three Swedish multilevel studies reported increased odds of cardiovascular disease, including coronary heart disease and stroke with availability to fast-food outlets.^29–31 However, most of the increased odds of cardiovascular disease were explained by neighborhood-level deprivation and individual-level socioeconomic status (SES), which was in agreement with a prior finding showing that the association of unfavorable food environments with cardiovascular risk profile was mainly driven by the area and individual income, rather than food accessibility.³² Conversely, subsequent to full adjustment for a broad range of individual-level and environmental covariates, the association of neighborhood-SES or fast-food density with prevalent cardiovascular diseases was fully attenuated in a Canada study using a multilevel approach.¹⁹ Moving beyond the neighborhood represented in terms of SES composition, our study focused on examining various types of ready-to-eat food environments which capture features of land use patterns for food outlets and access to food resources. Our consistent findings in the fully adjusted and restricted cubic spline models comprising the single-type and composite measure of all three types of ready-to-eat food environments in proximity and density indicated that the significant associations between food environments and HF risk was independent of the neighborhood socioeconomic disadvantage and other covariates.

Although the mechanisms related to the development of HF with accessibility to the food environment have yet to be determined, changes in individual-level biological and behavioral risk factors in response to the context of food environments might partly contributed to the association between food environments and HF. An unfavorable ready-to-eat food environment with great accessibility to pubs and bars or fast-food outlets might influence eating and drinking behavioral patterns that have been related to an elevated HF risk. The adverse food environments may deteriorate reliance on energy-dense and nutrition-poor food, soft drinks, alcohol, and other beverages,^33–35 leading to elevated harmful cardiovascular risk factors such as obesity^24,36–42 and hypertension ⁴³ and in turn, might increase HF risk. Further investigations are needed to explore the underlying mechanisms.

Interestingly, our observed relationships between ready-to-eat food environments and heightened risk of HF were further nuanced when considering the impact of other social determinants of health, which collectively shape an individual’s susceptibility to HF. Specifically, we found that the relations between ready-to-eat food environments and elevated HF risk were more pronounced in participants who were poorly educated and living in urban areas without formal physical activity facilities within 1 km of the residential address. These modification effects of education and urbanicity could be attributed to the socioeconomic disadvantage acting as a proxy for the above characteristics with a measurable influence on cardiovascular risk profiles.⁴⁴ These findings indicated that participants with socioeconomic disadvantages had a higher possibility to be affected by ready-to-eat food environments. The significant relationship between food environments and HF could be exacerbated by the presents of socioeconomic inequality. In addition, the associations between food environments and HF were more evident in residents with a lower density of neighborhood formal physical activity facilities. This finding is supported by the fact that low physical activity, which is influenced by the neighborhood physical activity environment, is a risk factor for HF.^45,46 Looking ahead, as highlighted in the scientific statement from American Heart Association, another notable social determinant contributing to risk of HF is food insecurity, characterized by limited or uncertain access to healthy foods.⁴⁷ Food insecurity is frequently interconnected with low SES composition and ready-to-eat food environments. To tackle the intertwined relationship between food insecurity, ready-to-eat food environments, and HF risk, it is imperative to conduct further investigations within other longitudinal cohorts encompassing comprehensive measurements of both specific food insecurity and the surrounding ready-to eat food environments.

The strengths of the study included the use of high-quality UK biobank data for the large sample size, geographical variability, and longitudinal long-term follow-up; objective evaluation of single-type and composite ready-to-eat food environments exposures at the individual level using metrics comprising both proximity and density; accounting for a wide range of lifestyle and environmental confounders; and a set of sensitivity analyses to validate our results. However, we acknowledged the current analysis had several potential limitations. First, food experimental exposure misclassification was possible since the residential neighborhoods do not necessarily represent the place in which they obtain and consume food. Further research is needed to account for the effects of food environments in participants’ working place. In addition, exposure misclassification may arise from movements between neighborhoods during the follow-up period. However, the stability of our sample with around 70% of residents living in the current address for at least 10 years would reduce the likelihood of this bias. Second, it’s important to consider the potential changes in ready-to-eat food environments over the 12-year follow-up period, which could introduce time-varying covariance. While employing a Cox regression model with valid assumption of proportional hazards can mitigate biases stemming from time-varying variations and enhance estimation accuracy, further investigation is advisable to delve into the dynamic nature of these food environments over time. Third, as in other observational studies, we could not rule out the risk of unadjusted residual neighborhood confounding that may coexist with food environments to affect HF risk, for example, social support, access to healthcare services, air pollution, and noise. Fourth, the absence of food insecurity data within our study was recognized as a limitation. Though food insecurity was a significant risk factor for cardiometabolic disease, it is often cited as an obvious, measurable marker of socioeconomic position. Its potential impact may be partly captured by the collective socioeconomic factors considered in our analysis, potentially constraining its influence on our findings. Additionally, food insecurity, closely linked with ready-to-eat food environments, leads individuals to resort to energy-dense, nutrient-poor foods due to economic constraints and limited access to healthier options. Adjusting for food insecurity might potentially lead to over-adjustment, given its inherent association with the core aspects of food environment measurements and their joint influence on incident HF risk. Further validation of our findings in other longitudinal cohorts equipped with specific assessments for both food insecurity and food environments could further enhance the robustness and generalizability of our conclusions. Finally, though UK biobank was a large and geographically diverse cohort, it may not represent the general population as a result of the restriction to white European ancestry, low response rate of 5.5%, and evidence of health volunteer bias.⁴⁸ This may limit the generalizability of our findings to other populations.

In conclusion, we found exposure to ready-to-eat food environments was related to a higher risk of incident HF. Our findings lend support to improvement of neighborhood food environments in the prevention of HF.

Short commentary:

a. What is new?

• Exposure to ready-to-eat food environments is related to a higher risk of incident HF, particularly for the types of pubs and bars and fast-food outlets.

• The associations between ready-to-eat food environments and HF risk are modified by education, urbanicity and density of physical activity facilities.

b. What are the clinical implications?

• Our findings indicate the potential importance of improving food environments in the prevention of HF.

• Improving educational attainment, urban infrastructure, and the availability of nearby physical activity facilities could substantially reduce the increased risk of heart failure linked to ready-to-eat food environments.

Abbreviations:

HF

Heart failure

UKBUMP

UK Biobank Urban Morphometric Platform

AHR

Adjusted hazard ratio

SES

Socioeconomic status